神经网络在图像压缩技术中的应用

Neural Network Apply to Image Compression

综述了神经网络作为图象压缩信号处理工具的方法。由于神经网络的大规模并行及其分布结构，使之成为解决数据压缩的有力工具；网络的特征与人类视觉系统的特征相类似，这就使我们更容易处理视觉信息。例如，多层感知可作为差分脉冲编码调制的非线性预测器，已证明这种预测器较线性预测器可改进预测效果。另一活跃的研究领域是用Ｈｅｂｂｉａｎ学习规则获取主分量，主分量是理想的线性ＫＬ变换的基向量。这些学习算法的计算更优越于传统特征分解技术并适应于输入信号的变化．还有另一种模型为ＳＯＦＭ（ｔｈｅＳｅｌｆ－ＯｒｇａｎｉｚｉｎｇＦｅａｆｕｒｅＭａｐ）已在向量量化码本设计中有许多成功的应用。结果码本与标准ＬＢＧ算法相比对初始条件的依赖程度较低，且码字的拓扑序可用来进一步提高编码的效率并减低计算的复杂性。

This paper presents methods of neural network as signal processing tools for image compression. They are well suited to the problem of image compresson due to their massively parallel and distributed architecture. Their characteristics are analogous to some of the features of our own visual system, which allow us to process visual information with much ease. For example, multlayer perceptions a can be used as nonlinear predictors in differential pulse code modultion(DPCM). Such predictors have been shown to incease the predictive gain relative to a linear predictor. Another active area of research is in the application of Hebbian learning to the extraction of principal components, which are the basis vectors for the optimal linear Karhunen Loeve transform (KLT). These learning algorithms are iterative, have some computational advantages over standard eigendecomposition techniques, and can be made to adapt to changes in the input signal. Yet another model the sellf organizing feature map (SOFM), have been used with a great deal of success in the design of codebooks for vector quantization (VQ). The resulting codebooks are less sensitive to initial conditions than standand LBG algorithm, and the topological ordering of the entries can be exploited to further increasiency and reduce computational complexity.